v3.0.0 — 22nd-Pass Audit Sync: Full Template Parity

Safety-Critical Gate Fixes

• 6V gate: Corrected to "worse-of" dual system (percentage-based AND fixability-based)
• Post-6R FAIL routing: Corrected to "fix 6V-C manually" (human-judgment issues)
• 6P max-cycle exhaustion: Expanded with "effectively converts to CONDITIONAL PASS" framing
• Stage 1 gate: Reclassified — no formal PASS/FAIL gate
• 6V/7V scenario caps: Split into distinct rows (150 for 6V full, 15 for 7V smoke)

Template Files (all synced byte-identical with canonical template)

• CLAUDE.md: Missing Stage 3 prerequisite, expanded 6R skip condition, 6D skip condition, Stage 8 WARN correction
• execution-guide.md: 13 differences resolved including 6P-R prerequisite precision, credential tiers format, Stage 8 batch limit removal
• feature_scenario_generation.md: 10 updates — Step 0 filter, generalized paths, cycle detection procedure, Mermaid validation, flaky handling
• All 6 rule templates: Cross-reference corrections, baseline rule count, EMPTY_COMPONENT comment, testing cross-reference

Skill Detailed Guides (all 23 replaced with canonical prompt files)

• All stage prompts now match the 22nd-pass audit versions exactly
• Stale duplicate reference files removed
• SKILL.md reference paths updated

Agent Definitions (all 7 updated)

• feature-tests: Narrowed to E2E-only
• feature-reviewer: Checklists aligned (10 items, blocking/secondary)
• feature-backend: Crash recovery, Session Language, 13 backend testing sub-rules
• feature-frontend: 7 frontend testing pitfalls, expanded design guidelines, i18n
• brd-writer: Fixed domain naming (3 mismatches), dynamic token budget
• prd-writer: Dynamic token budget, BRD issues handling
• transmute-pipeline: Stage 4 adds .claude/rules/*.md verification